Five geological criteria, one verdict, and a screening cycle that drops from roughly four months to under ten minutes per site. That is the shape of the opportunity when an agentic pipeline, a set of cooperating software agents that each own one expert check, takes over the first pass of CO2 storage suitability. The paradox of carbon capture and storage (CCS), the practice of injecting captured CO2 into deep rock formations for permanent disposal [1], is that the slowest and most expensive part of a storage programme is not injecting the CO2. It is deciding where to even look. Every figure in this briefing is illustrative, from a labeled synthetic demonstration; no real deployment is measured.
The bottleneck nobody budgets for
A CCS operator holding a portfolio of exploration licences faces a triage problem before an engineering problem. Each block must be checked on five criteria: depth, porosity and injectivity (how readily the rock accepts injected fluid), caprock seal integrity (whether the impermeable layer above the store actually holds), structural closure (whether the trap geometry contains the plume), and an induced-seismicity stress margin (how much injection the local stress state tolerates before triggering felt earthquakes [3]). Done traditionally, a small team of specialists works one site at a time, and a single screening pass can run three to four months (illustrative, synthetic demo). Across twenty blocks, the calendar, not the geology, becomes the constraint on how fast capital can be committed.
Five agents, one auditable verdict
EarthScan CrustScreen wraps that expert workflow in five cooperating agents, one per criterion, each returning a plain pass, marginal, or fail with the evidence behind it. A depth agent confirms the reservoir sits in the supercritical CO2 window, deep enough that CO2 stays in its dense, storage-efficient phase [2, 4] yet shallow enough to drill economically. A porosity agent scores injectivity, a seal agent tests the caprock, a closure agent checks trap geometry, and a seismicity agent estimates the stress budget against planned injection. The agents run in parallel and reconcile into a single GO / REVIEW / NO-GO verdict per block, rolled up into a portfolio-wide injectivity index.
What changes in the workflow
In the demonstration with a North Sea CCS operator, CrustScreen re-ran the full five-criterion pass in under ten minutes per site against a three-to-four-month manual baseline (illustrative, synthetic demo). Across a 24-block licence portfolio, an estimated 18 months of sequential specialist effort collapsed into a single afternoon of triage: all 24 sites screened, the 6 worth deeper study flagged, and an estimated 2.1 million dollars in early-stage evaluation cost avoided (illustrative, synthetic demo). The experts were not removed. They were redeployed from rote first-pass filtering onto the handful of blocks that actually warranted their judgment.
Permit-readiness as a by-product
Because every agent verdict ships with its inputs, thresholds and evidence trail, the output doubles as a permit-readiness artefact. A regulator or joint-venture partner can see exactly why a block scored GO, which is the difference between a screening result and a defensible screening result. In the demonstration, packaging that trail turned a normally bespoke reporting task into an automatic export, trimming an estimated further three weeks off the path to a storage-permit application (illustrative, synthetic demo).
The bigger picture
CCS at climate-relevant scale is a throughput problem. Thousands of candidate storage sites need appraisal this decade, and the pool of geoscientists qualified to screen them is finite. Agentic screening does not replace that expertise; it lets one expert's judgment be applied consistently across a hundred sites instead of exhausted on three. The strategic value is compressing the time between holding a licence and knowing whether it is worth developing, so capital chases the best acreage first and stranded appraisal spend falls across the industry.
Limitations
- Every number here is a labeled synthetic demonstration. No production deployment exists, and the timings, block counts and cost figures are illustrative, not measured field results.
- The five checks are a first-pass screen, not an appraisal. A GO verdict starts detailed site characterisation; it does not replace it.
- The agents' thresholds are schematic. Real screening criteria are basin-specific and regulator-specific.
- Expert oversight remains in the loop. The demonstration redeploys specialists; it does not remove them.
By the numbers
| Metric | Manual baseline | Agentic pass |
|---|---|---|
| Screening cycle per site | 3 to 4 months | under 10 minutes |
| 24-block portfolio | ~18 months of sequential effort | one afternoon |
| Blocks flagged worth deeper study | n/a | 6 of 24 |
| Early-stage evaluation cost avoided | baseline | ~$2.1M |
| Path to permit application | bespoke reporting | evidence trail exported, ~3 weeks saved |
All figures are illustrative synthetic-demo values, not measured field results.
Key takeaways
- The gate is triage, not injection: the slowest, costliest phase of a CCS programme is deciding where to look.
- Five criteria, one auditable verdict: depth, injectivity, caprock seal, structural closure and seismicity margin each get an agent, and each verdict carries its evidence.
- Months to minutes (illustrative, synthetic demo): a three-to-four-month manual pass compresses to under ten minutes per site, so a whole portfolio is triaged in an afternoon.
- Experts are redeployed, not replaced: specialists move off rote first-pass filtering and onto the few blocks that genuinely need judgment.
- The output is permit-ready: because every check ships with its inputs and thresholds, screening results arrive as defensible, exportable evidence.
References
[1] Carbon capture and storage (overview): Wikipedia
[2] Supercritical carbon dioxide (dense-phase storage window): Wikipedia
[3] Induced seismicity (injection-triggered earthquakes): Wikipedia
[4] IPCC. Special Report on Carbon Dioxide Capture and Storage. Cambridge University Press, 2005. ipcc.ch
[5] Source: internal briefing note and its accompanying screening-rig figure, whose base map follows the 2018 PhD thesis, Fig 3.2 (study-area map). The agents, thresholds and portfolio outcome are a labeled synthetic-demo illustration, not measured results.
EarthScan builds agentic subsurface screening for teams deciding where storage capital goes next. If your licence portfolio is waiting on a calendar instead of a verdict, we should talk.
